ISSN: 2277-9655 [Yawale* et al., 6(1): January, 2017] Impact Factor: 4.116 IC™ Value: 3.00 CODEN: IJESS7

IJESRT INTERNATIONAL JOURNAL OF ENGINEERING SCIENCES & RESEARCH TECHNOLOGY PEDESTRAIN DETECTION BY VIDEO PROCESSING USING THERMAL AND SYSTEM Ankit Dilip Yawale*, Prof. V. B. Raskar * Department of Electronics and Telecommunication JSPM’s Imperial College of Engineering and Research Pune, India Department of Electronics and Telecommunication JSPM’s Imperial College of Engineering and Research Pune, India

DOI:

ABSTRACT The paper describes the use of thermal camera and IR night vision system for the detection of Pedestrians and objects that may cause accident at night time. As per the survey most of the accidents cause is due to low vision ability of human at night time, which leads to most dangerous and higher number of accidents at night with respect to day time. This system include the IR night vision camera which detects the object with the help of IR LED and photodiode pair, this camera have capability to detect the object up to 100m. The thermal camera detects the heat generated by any of the object like , Human animals etc. which gives us the facility to detect the object for higher range and with low reflective surface where IR night vision may fails. With the use of these two cameras mounted on which helps the driver to drive safely. In this system, HOG (Histogram of orientated gradients) algorithm and support vector machine (SVM) is performed with the help of OpenCV in Matlab and EmguCV in Visual Basic 2012. The system is tested on the video recorded using these cameras, and got good and efficient result. And this system is cost efficient and easy to implement.

KEYWORDS: Thermal and IR Night vision, OpenCV, Video Processing, Object and Human Detection, Automobile Safty.

INTRODUCTION Humans are the most intelligent creation of god. Human’s one of the most important discover is transportation systems. But with the increase of population, needs and desires of human being road transportation is getting bit one of the major reasons of human made death. Pedestrian Detection system is now requirement in the new era of transport. Is ancient time we use to travel by bullock cart or on a horse which does not causes any causality. But now-a-days people have 2-3 vehicles at their house which increases the traffic at roads and accidents and eventually deaths cause by it.

In this system, we have tried to reduce the rate of accidents by more than a half. This project help to build a good and efficient and cost affective system for car/ automobile manufacturers to develop more safe and luxuries vehicles. As the population and needs of human being is increasing the numbers are increasing government is trying to decrease the rate of causality and death.

In recent years, According to survey 38% of fatal accidents in the European Union occur in darkness places, instead the fact that, the traffic during nights time is several times smaller than that on a day. This means that the risk of an accident in darkness increased. This stats has a strong relation and effect on the pedestrians. Currently, the pedestrians are of about 20 % of all traffic accidents. More than half of pedestrian deaths take place at night (51 %).

As per survey by The Times of India, in various major city of India (figure 1),we can see that 60% of accident occur at Delhi is at night time. So it is more necessary to find a good solution for reducing the numbers.

http: // www.ijesrt.com © International Journal of Engineering Sciences & Research Technology [29] ISSN: 2277-9655 [Yawale* et al., 6(1): January, 2017] Impact Factor: 4.116 IC™ Value: 3.00 CODEN: IJESS7

Fig. 1. Accident statistics of Delhi city with time in 2015

This system will assist the driver to see much better in darkness and able him to drive safely. This system provides him a that will show him the live video of road and also continually alerting him about the pedestrians and handles in the way. This system can also assist drive who got asleep while long drive by flashing an alert at critical case

THERMAL AND NIGHT VISION SYSTEM Thermal Vision System Thermal vision system uses thermo-graphic camera or thermal imaging camera. This camera works same as the normal visible light camera but this camera uses radiations. This camera operates at very long wavelength of nearly 14,000 nm (or say 14 µm) instead of visible light range of 400-700 nm. Infrared energy is just one part of the electromagnetic spectrum, which encompasses radiation from gamma rays, x-rays, ultra violet, a thin region of visible light, infrared, terahertz waves, microwaves, and radio waves. These are all related and differentiated in the length of their wave (wavelength). All objects emit a certain amount of black body radiation as a function of their temperatures.

The higher an object's temperature, the more infrared radiation is emitted as black-body radiation. A special camera can detect this radiation in a way similar to the way an ordinary camera detects visible light. It works even in total darkness because ambient light level does not matter. This makes it useful for rescue operations in smoke- filled buildings and underground.

Fig. 2. A thermal image showing temperature variation in a hot air balloon

http: // www.ijesrt.com © International Journal of Engineering Sciences & Research Technology [30] ISSN: 2277-9655 [Yawale* et al., 6(1): January, 2017] Impact Factor: 4.116 IC™ Value: 3.00 CODEN: IJESS7 Thermal camera is capable of detecting the object normally up to 200-300m, which very useful for our system. This camera can be used to sense the human as well as radiations emitted by cars or vehicle. In this band spectrum the human body which usually have temperature of about 300K and that has the highest energy emission, therefore objects with an internal heat source, such as pedestrians, cars in motion (e.g., engine, radiator, heated reflectors) are clearly visible.

IR Night Vision System In this system, we have to use a pair of IR illuminators and Near Infrared camera. IR illuminator illuminates the IR range light which reflects back and captured by the near infrared camera. this type of system is usually used in the ATMs, CCTV surveillance systems. This allows the user to see in low light, where the human vision cant able to see properly.

This system is capable of detecting the object nearly up to 100 to 150m. These cameras are less in cost than that of thermal vision cameras. We can use simply night vision system for less cost solutions and for hi-tech solution we will use the both thermal and infrared night vision system.

Fig. 3. A Infrared night vision image showing car and pedestrains

This system gives a best result for the pedestrians and object detection system.

Existing solution at automobile manufactuer The first night vision system in the automobiles has been introduced to the market bya company was General Motors in the year 2000 and it is applied in the Cadillac DeVille. Development of this project took 15 years of 70 person’s team and costed approximately $100 million. After this in 2003, Toyota has firstly develops the commercial grade active night vision system for the car Toyota Landcruiser and LX470 which can reached up to the range of 100 m. In 2004 Honda has develops the same in the Legend model, which was an optional system named as "Intelligent Night Vision" with the prime option as pedestrian detection. The system is capable of the range between 30 and 80 m .Now-a-days the major automobile manufacturing companies are taking most efforts on the safer and luxury vehicle. They are investing more on safety of the driver. Audi, BMW and Mercedes- Benz are at the top in this race.

Table i. Evolution of night vision systems Active systems Date Passive Systems 2000 “Night vision”, General Motors, Cadillas DeVille "Night View", 2003 "Intelligent Night Toyota 2004 Vision System",

http: // www.ijesrt.com © International Journal of Engineering Sciences & Research Technology [31] ISSN: 2277-9655 [Yawale* et al., 6(1): January, 2017] Impact Factor: 4.116 IC™ Value: 3.00 CODEN: IJESS7 Landcruiser, , Lexus X470 Pedestrian detection "Night View 2005 BMW Night Assist Plus", Vision", BMW 7 Mercedes-Benz series S-Class "Night View 2006 “BMW Night Assist Plus", 2007 Vision", BMW 5, Mercedes-Benz 6 series CL-Class "Night View", 2008 BMW Night Vision", BMW 7 Hybrid, series, Pedestrian pedestrian detection detection "Night View", 2009 "Night Vision Lexus LS, 2010 Assistant", Audi pedestrian A8, Pedestrian detection "Night detection View Assist Plus", Mercedes- Benz E, S-Class Pedestrian detection "Night View 2013 BMW Night Assist Plus", 2014 Vision", BMW 7 Mercedes-Benz series, Pedestrian S-Class, Animals detection detection

PROPOSED SYSTEM OF NIGHT VISION In this design, author has used two cameras for detection of the pedestrians. Author has suggested two type of system to be implemented. First (High-quality system), high cost solution which consists of Thermal camera as well as IR night vision camera. And the second (Low-quality system) is having only IR near night vision camera, this system has low cost than that of first one but we have to scarify with the range. First system is having range nearly as 200-250m and second system have 100-125m of range.

Fig. 4. Camera placement of high-quality night vision system

High-quality Night vision System This system consists of both thermal vision system (i.e. Passive system) and Near IR camera system (i.e. Active system). This system has the best performance as both the cameras compensate weaknesses. High-quality system consists of a thermal imaging camera which is placed on the radiator grille, the thermal camera cannot be placed behind the or radiator grille because it will detect the heat or the radiation produced by its own

http: // www.ijesrt.com © International Journal of Engineering Sciences & Research Technology [32] ISSN: 2277-9655 [Yawale* et al., 6(1): January, 2017] Impact Factor: 4.116 IC™ Value: 3.00 CODEN: IJESS7 vehicle.it is a IR filtering property of Thermal cameras. Near IR night vision camera is placed behind the windshield of the vehicle front of the rearview mirror which allows driver to get better view of the road. IR illuminators are placed in the headlights of the car. As the Near IR camera is placed up at the vehicle, it gives wide angle but the distance range in less. But the thermal camera or the passive system have the long range but narrow angle. In this way, both of these camera compensates each other’s weaknesses.

Fig. 5. Assumed pedestrian detection range

Low-quality Night vision system The low-quality system has almost the same design as high-quality system. But the difference is that it is limited to near IR night vision system i.e. active system only. In system, cost is reduced by removing the costly thermal imager. Pedestrian detection range for this system is 100-125 m.

Cost of both the systems The cost of the system is one of the most important parameter after Range efficiency and capture angle. The cost includes the cost of each components use to develop the system.

This systems requires mainly central control unit i.e. processor, two cameras, LCD, IR illuminator, and wiring. This work only deals with software part. But the proposed hardware for building this system are:- Beagle Bone (Sitara 35XX Processor), LCD, Thermal Camera ( FLIR Tau 2-336) and IR Illuminator. Costing for these are as follow:-

Table ii. Costing of component used to develop the system Component High- Low- Quality Quality System System Beagle Bone (Sitara $65 $65 35XX Processor) LCD $100 $100 Near IR Camera $150 $150 IR Illuminator $200 $200 Thermal Camera ( $2,750 N.A. FLIR Tau 2-336) Other Accessory $85 $85 Total $600 $3,350

IMPLIMENTATION OF THE SYSTEM People and object detection is most current topic in the digital image processing. Many new algorithms and techniques are developed in recent few years. In maximum implementation basic on this are either with passive or with active night vision system. Many of those are based on Support vector machine (SVM) or BLOB or neural network and others.

In this implementation author has used Histogram oriented gradient (HOG) algorithm using OpenCV library. Firstly the image is acquired from cameras. After acquiring the image, it is processed. The sequence of processing is as follows.

In the pre-processing stage the blur caused due to moving object with respect to vehicle is removed. This gives us

http: // www.ijesrt.com © International Journal of Engineering Sciences & Research Technology [33] ISSN: 2277-9655 [Yawale* et al., 6(1): January, 2017] Impact Factor: 4.116 IC™ Value: 3.00 CODEN: IJESS7 process able image. After this step we have to prepare our region of interest (ROI). This is very important step, if we select incomplete or improper region of interest then the result of that frame will not be proper. In other words, if at this step of region of interest misses the object of pedestrians then this frame is waste. First step in ROI is segmentation; segmentation is a process to abstract the desired region from the image background. Usually we use threshold maps in normal application and disparity maps in stereovision systems. In this work a modified dual-threshold adaptive threshold [12] was used. The algorithm converts the input gray-scale image to a binary image, where white objects are the potential candidates and the background is black (see Fig. 7). It works adaptively under various lighting conditions and the contrast level.

Fig. 6. Flowchart of processing

Fig. 7. Output of segmentation and threshold segmentation of a thermal camera image: From Left upper row: Original image, Gray scale image, threshold gray scale : from Left Lower row :binary image, segmented image, Threshold segmented image

The additional threshold segmentation gives a better result to work further. After the image is segmented, the process of morphological opening removes small artifacts. Next process on the binary image is to select the interconnected group of pixels which maybe the object we are trying to detect and then it is labeled and it’s all properties like length, width height are measured so that we can assume what actually the image is, This process known as connected component labeling (CCL).

As we got the region of interest and all other details about it. It is very crucial that the quality of image we got for the final stages is good or not, is it contents the required object with proper gradients. The quality of this image directly affects the result of object and pedestrians detection. First step here at the final stage of processing is feature extraction. This is done by reducing the unwanted data from an image. After this we will apply Histogram

http: // www.ijesrt.com © International Journal of Engineering Sciences & Research Technology [34] ISSN: 2277-9655 [Yawale* et al., 6(1): January, 2017] Impact Factor: 4.116 IC™ Value: 3.00 CODEN: IJESS7 oriented gradient (HOG) algorithm which helps us to detect what actually the object is. Is it the one we trying to find or not. And it will mark the object which has been detected.

The last stage that finally validates the object is a classifier. The most common classifiers are: support vector machine (SVM) as example of the supervised learning method, neural networks, self-organizing maps (SOM), and matrices of neurons [15]. A very helpful algorithm during classification is the boosting algorithm.

To summarize, in the presented system the following solutions have been applied:  modified adaptive dual-threshold for the image segmentation  Connected Component Llabeling (CCL) for selection of candidates  Histogram of Oriented Gradients (HOG) for feature extraction  Support vector machine (SVM) for training of the classifier.

RESULT AND EXPERIMENT OF THE SYSTEM The proposed Pedestrian Detection by Video Processing using Night thermal Vision System where the use of Matlab 2014(B) version and the Image acquisition and Computer vision systems matlab package is used to do the image processing.

Firstly we have to load the image or video frame. As it is a night vision camera image the image will looks like it’s a grey scale image but it will not. So we have to convert the image is to grey scale to do processing on it. Then convert then grey scale image in to threshold binary image which is used to detect the corner using FAST object/ corner detection algorithm and point out all the analysed corners ( In the below images it is represented by red color *).

Then we have applied Extract histogram of oriented gradients (HOG) features to get the proper detecting of object in the image.

The below steps to anaylse the performance of the proposed system are carried out

Figure 7. Original output of night vision camera image(Top left), Grey scale image (top right), Threshold binary image (bottom left), simple binary (bottom right).

http: // www.ijesrt.com © International Journal of Engineering Sciences & Research Technology [35] ISSN: 2277-9655 [Yawale* et al., 6(1): January, 2017] Impact Factor: 4.116 IC™ Value: 3.00 CODEN: IJESS7

Figure 8. Corner Detection using FAST algorithm

Figure 9. Feature Extraction using HOG Algorithm.

Figure 10. Original output of night vision camera image(Top left), Grey scale image (top right), Threshold binary image (bottom left), simple binary (bottom right).

http: // www.ijesrt.com © International Journal of Engineering Sciences & Research Technology [36] ISSN: 2277-9655 [Yawale* et al., 6(1): January, 2017] Impact Factor: 4.116 IC™ Value: 3.00 CODEN: IJESS7

Figure 11. Corner Detection using FAST algorithm

Figure 12. Feature Extraction using HOG Algorithm

TABLE I. Figure 13. Original output of thermal camera image(Top left), Grey scale image (top right), Threshold binary image (bottom left), simple binary (bottom right).

http: // www.ijesrt.com © International Journal of Engineering Sciences & Research Technology [37] ISSN: 2277-9655 [Yawale* et al., 6(1): January, 2017] Impact Factor: 4.116 IC™ Value: 3.00 CODEN: IJESS7

TABLE II. Figure 14. Corner Detection using FAST algorithm TABLE III.

TABLE IV. Figure 15. Feature Extraction using HOG Algorithm

REFERENCES [1] Karol Piniarski, Pawel Pawlowski, Adam D.browski,Pedestrian Detection by Video Processing in Automotive Night Vision system, SIGNAL PROCESSING ALGORITHMS, ARCHITECTURES, ARRANGEMENTS, AND APPLICATIONS,SPA 2014, September 22-24"', 2014,pp. 104-109. [2] J. Ge, , Y. Luo, G. Tei, Real Time Pedestrian Detection and Tracking at Night time for Driver-Assistance Systems, IEEE Transactions on Intelligent Transportation Systems, Vol. 1 0, No.2, 2009, pp. 283 -298 [3] N. Dalal, B. Triggs, Histograms of Oriented Gradients for Human Detection, IEEE Conference on Computer Vision and Pattern Recognition, ( Vol. 1 ).2005 [4] Honda, Honda Develops World's First Intelligent Night Vision System Able to Detect Pedestrians and Provide Driver Cautions, 2004, http://world.honda.com/news/2004/4040824_01 /video/index.htm I. [5] European Commission, Towards a European road safety area: policy orientations on road safety 201 1 - 2020, Brussels, COM(20 I 0) 389, 20 1 0. [6] Open source computer vision - OpenCV, hltp://opencv.org, 201 4

http: // www.ijesrt.com © International Journal of Engineering Sciences & Research Technology [38]